Gemini 3 release dates: the verified timeline, through 3.5.

Quick answer: the verified timeline.

Google announced Gemini 3 Pro and the Gemini 3 Deep Think mode on November 18, 2025, opening the Gemini 3 family. Gemini 3 Flash followed on December 17, 2025, and immediately became the default model in the consumer Gemini app and in Search's AI responses. Gemini 3.1 Pro shipped as a preview on February 19, 2026, per Google's own blog and the Vertex AI docs. Gemini 3.5 Flash arrived on May 19, 2026, announced at Google I/O as the first release of the 3.5 generation, positioned as Google's strongest model for coding and agents. That is everything released as of mid-July 2026. The missing family member is Gemini 3.5 Pro: Google said at I/O that it was in internal use and expected shortly, and subsequent press reported the date slipping into July; it had not shipped at the time this page was checked. Both released 3.x ids in the catalog here, gemini-3.1-pro-preview and gemini-3.5-flash, are callable today through any OpenAI-compatible client.

The Gemini 3 family history, release by release.

The cadence tells its own story: Google has shipped a major Gemini release roughly every two to three months since late 2025, alternating between Pro-class reasoning models and Flash-class speed models. Gemini 3 Pro launched with the Deep Think reasoning mode and Google's antigravity push into agentic coding. Gemini 3 Flash was the volume play, fast enough and priced low enough to become the consumer default within a week of launch. Gemini 3.1 Pro was pitched at complex tasks: repository-scale code comprehension and multi-source reasoning over its 1M-token window. Gemini 3.5 Flash inverted the usual order by leading the 3.5 generation with the Flash tier; Google's launch materials claim it outperforms 3.1 Pro on coding and agentic benchmarks like Terminal-Bench while generating several times faster, which is a vendor claim worth testing on your own workload rather than taking on faith.

Announced vs rumored: keeping the 3.5 Pro question honest.

What Google has actually said: at I/O in May 2026 the company stated Gemini 3.5 Pro was already in internal use and that it looked forward to rolling it out in the following month. That is an announcement of intent with a soft window, not a dated commitment. What the press has added: reports in early summer described the release slipping to July 2026 for quality refinement after enterprise testing. Plausible, sourced to coverage rather than to Google, and already past its implied window once; treat any specific 3.5 Pro date you read as reporting, not vendor fact, until Google publishes one. What nobody serious has claimed: dates for anything beyond 3.5 Pro. The family's two-to-three-month cadence invites extrapolation toward a Gemini 4, and extrapolation is all it is. The practical hedge is the usual one: integrate against exact released ids behind an endpoint that serves several families, and a future release becomes a one-string A/B rather than an event you wait on.

How to call the released Gemini 3.x models today.

Both released ids are addressable now through any OpenAI-compatible client: base URL https://api.apisrouter.com/v1, one key, model gemini-3.1-pro-preview for reasoning-heavy work or gemini-3.5-flash for speed and agent loops. No Google Cloud project, AI Studio account, or Vertex setup is required through the gateway, and the same key runs the GPT, Claude, and DeepSeek ids you would compare them against. Note the id spelling: the 3.1 Pro entry carries its -preview suffix, matching how Google shipped it, so pin the exact string from /v1/models rather than guessing at a stable name. Both catalog entries list a 1M-token context window here; budget prompts against the catalog figure as called from this endpoint.

import os
from openai import OpenAI

client = OpenAI(
    api_key=os.environ["APISROUTER_API_KEY"],
    base_url="https://api.apisrouter.com/v1",
)

for model in ("gemini-3.5-flash", "gemini-3.1-pro-preview"):
    resp = client.chat.completions.create(
        model=model,
        messages=[{"role": "user", "content": "One sentence: what did Gemini 3.5 Flash ship for?"}],
        max_tokens=100,
    )
    print(model, "->", resp.choices[0].message.content)

Pay-as-you-go · transparent per-model pricing

Selected models are priced below official list prices. Exact input, output, cache, and per-request prices are shown for each model.

Choosing between 3.1 Pro and 3.5 Flash while 3.5 Pro is pending.

Whichever way the test comes out, the integration does not change: both ids live behind the same /v1 surface, so the choice stays a config value rather than an architecture.

• Default to gemini-3.5-flash for agents, coding loops, and anything latency-sensitive: it is the newer generation, Google's own benchmarks place it above 3.1 Pro on agentic work, and it carries the lower catalog rate of the two.
• Keep gemini-3.1-pro-preview for deliberate long-context reasoning: multi-document analysis and repository-scale comprehension were its launch positioning, and generation-over-generation claims from vendors do not always transfer to every workload shape.
• Do not park work waiting for 3.5 Pro: it has no confirmed date, and a pending model is not a plan. Ship on a released id and make the upgrade a one-string eval when it lands.
• Run the comparison rather than reading it: the same prompt against both ids, plus a GPT or Claude id as an outside reference, settles in an afternoon what launch posts argue about for weeks.

Why Gemini release dates confuse people more than most.

Three Gemini-specific patterns feed the confusion. First, app-versus-API staging: Google routinely ships a model into the consumer Gemini app and Search before or after the API, so "it is out" and "I can call it" have different dates. Second, preview suffixes: 3.1 Pro shipped and stayed in preview naming, which reads as unreleased to people scanning for a stable id even though it serves production traffic. Third, tier leapfrog: the 3.5 generation led with Flash rather than Pro, so a search for the 3.5 release date finds a released Flash and an unreleased Pro at the same time, and both answers are correct for different models. The reliable check beats all of it: list the models your endpoint serves and look for the exact id. If it appears in /v1/models, it is released for you; everything else is press-cycle detail.

curl -s https://api.apisrouter.com/v1/models \
  -H "Authorization: Bearer $APISROUTER_API_KEY" | grep gemini

FAQ